Heatsink Testing Procedure

[splat]

Since we don't have the luxury of having one specific test station to do all our heatsink reviews on, here is a standard procedure to ensure everyone is obtaining their results in a similar, scientific manner.

This is just a preliminary draft to open the discussion. Please comment and make suggestions so we can fine tune this.


Place a pea sized (1/2 pea) spot of Thermal Interface Material (TIM) on the processor. If you are only using one type of TIM for all tests, then use the manufacturer's recommended application method. Some heatsinks may require applying a small amount to the base first to ensure good coverage. State your method when you write your article.
Take a picture of the TIM before mount.
Install heatsink per manufacturers directions
Turn on computer, enter in to BIOS, make sure the heat sink fan control is off so the fan spins at its maximum speed.
When you reach the Windows desktop, open CoreTemp (http://www.alcpu.com/CoreTemp/), turn on logging, and let it sit for 10 minutes to record the idle temperature.
Open CPU-Z (http://www.cpuid.com/cpuz.php) and take a screenshot to record hardware config and CPU Voltage.
Use speedfan (http://www.almico.com/speedfan.php) or hwmonitor (http://www.cpuid.com/hwmonitor.php) to record the fan rpm and take a screenshot
Place a digital thermometer near the fan inlet on the heatsink to record the inlet air temperature. This should be checked and recorded at 30sec intervals. Inlet temp will affect results and needs to be accounted for.
Open Prime95 (http://www.mersenne.org/freesoft/), on the torture test dialog window, select Small FFTs and make sure the number of threads is correct
Start logging on CoreTemp
Hit Start on Prime95 and let it run for 15 minutes
Stop the logging in CoreTemp
Turn off the computer, remove the heatsink
Take a picture of the spread pattern of the TIM on both the processor and heatsink
clean the existing TIM off the heatsink and processor
Repeat the entire procedure two more times for a total of 3 full mounts/remounts.


in the place of Prime95 and CoreTemp, you can use OCCT (http://www.ocbase.com/perestroika_en/index.php?Download). under "Options", select "Generate CSV Files". Then Select the CPU: OCCT tab, under Test Type select Custom, have it run for 15 minutes, use the Small Data Set and Normal or High Priority.

after the 3 runs are done:

Import the .csv files in to Excel, calculate the mean (average), median, mode, and variance for each run. (idle and load)
If results between the three runs seem askew, then do two more mounts and throw out the high and low.
test at stock speed/voltage and find max overclock


ideas for other tests to enrich your article:

test multiple fan speeds (fan off, 10%, 50%, 75%, 100%)
test different TIM application methods (dot vs. line)

[MattNo5ss]

That's a great start.

A few comments:

Is this just for testing a single heatsink? If testing/comparing multiple heatsinks, then the program used, time loaded, and fan speed used are irrelevant as long as they are consistent between tests.

IMO, HDT heatsinks should have TIM placed in the gaps between heatpipes before the "pea" method is used.

[splat]

btw, I threw in IntelBurnTest just as one option. I know Prime95/Orthos have been the norm for years and everyone has their own personal program of preference, but is there really a difference when the goal is generating processor heat, not necessarily overall system stability?

MattNo5, i'd be interested to see hard numbers on if applying to the base does make a difference because I've been using the pea method and it appears to give great coverage, even around the pipes.

[MattNo5ss]

No, load program doesn't really matter for testing single heatsinks either. Just if you wanted this to be a true standard, then everything would have to be the same so results can be cross referenced between multiple reviews. (which is basically impossible with all the "reviewer"/hardware/fan/TIM/etc. options available)

When I had my HDT-s1283 I tried both mount types. I don't have the heatsink anymore, so I don't have hard numbers for you, but here's my experiences:

If the gaps weren't filled, then my spread patterns wouldn't reach as far across the heatpipes. This is b/c the TIM has to go up into the gaps and then backdown across another hetapipe, and repeat across the all the pipes. The gaps would "slow down" the spread from the heatsink pressure.

So, if you wanted to only do the "pea" method, then you would need more TIM on a HDT than a flat base, because the gaps need to be filled. On the other hand, filling the gaps first allows you to use the same "pea size" on both HDT and flat bases.

Side Note:
My Thermalright VX article won't be able to follow this procedure since my HDT has been sold. But, if I do future heatsink reviews I'll be able to follow a standard.

[MIAHALLEN]

[LIST=1]
Place a pea sized spot of Thermal Interface Material (TIM) on the processor.....

Place a digital thermometer near the fan inlet on the heatsink to record the inlet air temperature. This should be checked and recorded frequently (once a minute?) to ensure the inlet temperature is constant.

Pea size is too big IMHO....should only be 1/2 or less of a pea ;)

Inlet temprature should not be checked and recorded frequently (once a minute?) to ensure the inlet temperature is constant....because it will not stay constant. Instead, it should be checked regularly (I use about a 30 sec interval) and averaged for the run. Over ten minutes you might be surprized...I've seen 5+ degree swings in a single run as the heater in the room cycles.

btw, I threw in IntelBurnTest just as one option. I know Prime95/Orthos have been the norm for years and everyone has their own personal program of preference, but is there really a difference when the goal is generating processor heat, not necessarily overall system stability?
Open IntelBurnTest, set to Maximum and 5 runs.
The only problems with this is the test length will vary depending on the amount of installed memory....like if I only had a single 1GB DIMM installed, the test would not be long enough to generate a good heat load. We need a timed test if we want to standardize this :thup:

[EarthDog]

TIM application should be by the TIM mfg specs, no? That way no matter who does it with what TIM, the application amount/method would be the same.

Otherwise, looks good to me.

[splat]

updated the procedure a little.

[Bobnova]

HDT coolers absolutely need a different application style.
The difference with Perihelion for instance is 4-5*c between manufacturer's method (5.5mm blob) and my method (stripes down the pipes).

[bing]

Minor suggestion but imo its important, photo shoots, not mandatory though, but c'mon, in these days even crappy phone has it. This will light up the test results. ;)

For multiple HS/block tests or different TIM tests for example, shots of the TIM spreads after the dismounts will be valuable too if the tester's camera capable of macro shot.

[splat]

good call, "take a picture of the spread pattern" added.

btw, hokie suggested an hour as the run time for a stress test. I think that is sort of excessive for an air-cooling heatsink test...especially if we are mandating that a reviewer take manual air temp readings at 30sec intervals. Is 1 hr necessary or do we think we can cut that down and still properly load the heatsink and still get enough data points for a good test?

[thideras]

I do have a suggestion for higher temperatures, CoreDamage. I found this program a long time ago and I can no longer find it anywhere else or the original site. File is clean and it puts out a TON of heat. The only one that can match the heat output is Linpack. Considering the setup time for Linpack, this is a literal click and go program. Also, if the system is unstable, it keeps chugging; it does NOT do any system checks and will not stop until you close the program. I've hosted it on my website for easy access.

http://thideras.com/downloads/CoreDamage.exe

This program shut down my Lenovo Thinkpad in 30 seconds due to temps. My old laptop lasted 15 seconds :p

[QuietIce]

I think an hour-long run for air cooling is not needed. In 15 minutes the system is more than loaded and it's plenty of time for data collection if using 15s intervals (which is what I'd suggest). 2-3 minutes for loading the heatsink then 50 samples should be enough.

As for heat stressing I think Prime95 or OCCT should be used since we often recommend those for stress testing and determining load core temp.

HDT coolers absolutely need a different application style.
The difference with Perihelion for instance is 4-5*c between manufacturer's method (5.5mm blob) and my method (stripes down the pipes). I agree completely - (thin) stripes down the pipes is without a doubt the best method for those sinks! :)


Someone did a rough test on application methods for various heatsinks, though I can't recall who that was. I'm sure a Google search would find it for those interested ...

[bing]

Another one, apart from temperature, fan's rotation speed must be clearly stated during the test. So using fan without rpm reporting capability is not recommended.

[hokiealumnus]

Couple things to add. I don't think the stress testing program needs to be mandatory. I use Prime95 small FFTs to stress my CPU. Not because it's the hottest (see core damage / linpack - IBT (linky (http://downloads.guru3d.com/IntelBurnTest-v2.3-download-2047.html)) is a GUI for linpack and takes no time to set up btw), but because I like it. It's a personal preference. As long as the program stresses all cores and you use it consistently across all tests, it's good to go IMO.

To that effect, I would remove orthos from the list in the OP. Unless there has been an update, it does not stress more than two threads. Prime95, IBT (set to the correct number of threads), linpack, core damage - they'll all stress all threads.

An hour is entirely too long for an air heatsink. I personally would do no less than 10min; they do heat up over a minute or two and it's worth a few minutes extra to ensure you have accurate readings.

The reason I use an hour for water testing is because I have a lot of it - reservoir, two radiators & what's in the tubing/block. It takes a while to heat up.

[splat]

I had intel burn test in the guide first but I decided to change it back to orthos. I think prime95 or orthos (which uses prime95) set to small ffts is the best way to test the cpu because I've been using it for years for stability testing and it's never let me down. I know there have been other programs that have come along that may or may not create more heat but i know if i've tested stable with prime95 then i was always stable for folding or other tasks. So, I think prime95/orthos give good real world usage results.

if some one wants to write an article comparing all of the programs out there for stressing the CPU and find another one really is superior then maybe we'll change it. if some one really has objections to using prime95/orthos then all i can say is make sure you use the exact same procedure for all tests in your article.

edit: oh right, orthos is limited to 2 threads. forgot.

[hokiealumnus]

I'm all about Prime95. I don't think it should be a requirement as long as the person testing uses the same program throughout, but I'd not have anything against it being the program of choice in the OP. Orthos just doesn't do what we need any more.

[splat]

I'm all about Prime95. I don't think it should be a requirement as long as the person testing uses the same program throughout, but I'd not have anything against it being the program of choice in the OP. Orthos just doesn't do what we need any more.

i did an edit while you were posting this: i think having one set program is useful in case we try to make an all encompassing table similar to Joe's old heatsink table. Sure they'll all be tested using different processors as everyone will use their own hardware, but if we get rid of as many variables as possible then the diff in T results will be more reliable.

IBT i think we can forget because it does cycles instead of a steady stress. Prime95 itself is a little cumbersome to use simply for stresstesting, ORTHOS was a nice simple frontend. is there another frontend that is newer?

[hokiealumnus]

Edit doesn't show anything, still Orthos in there, FYI.

OCCT uses the prime95 engine (http://www.ocbase.com/perestroika_en/index.php) (and others, if you choose).

[QuietIce]

I like OCCT simply because it shows temps as well, no other program needed. Prime95 is also an excellent choice because of it's prominence in the OC'ing community as a whole.

The reason I use an hour for water testing is because I have a lot of it - reservoir, two radiators & what's in the tubing/block. It takes a while to heat up. I agree water cooling takes longer. My loop takes somewhere between 20-30 minutes just to get the temps stable. It's a big loop ... ;)

[splat]

if occt uses prime95 thats good enough for me. I'm assuming their "cpu" option is the small ffts.

edit: ok played with it a little. I think i still like normal prime95 better, but it looks like a good option, i updated the first post. I have always used the Small FFTs as I always understood that to strictly test the CPU for errors even tho the Large FFTs says "Maximum Heat". I always understood that to mean it tested the entire system instead of strictly the cpu. Small FFTs runs the cpu at 100% and I don't think there is really any way to get above 100% but i'm sure some one could straighten me out on that subject.

[splat]

do we want to require a picture of the spread pattern from every mount in the article? or is one enough? I think one would be enough.

[hokiealumnus]

I think one's enough as well.

[MattNo5ss]

If there are temp differences outside the margin of error between two or more mounts, then I think all the mounts should be shown as a possible cause behind the temp differences.

As long as all the mounts have about the same temps, then just one is fine.

[thideras]

There should be no "if this, then this". I would suggest taking pictures of all the mounts, regardless of the temperatures you get. You can then correlate spread patterns to temperatures, especially if one is out of the normal (higher or lower). The TIM spread is just as important to record as the ambient temperatures.

[bing]

Yeah, agree with Thiddy, also almost forgot, the shot of how the TIM was applied "BEFORE" mounting, I guess this is important as well looking at different HS bases like solid vs HDT type with those pipes.

[QuietIce]

I can understand taking the pics for record keeping purposes but I don't see a need to publish them with the main article. They could be attached to a post in the discussion thread. But if the patterns and temps are consistent it seems like one representational pic should be enough. No harm in posting them all, though, I guess - just seems like overkill if they're all the same ... :shrug:

[hokiealumnus]

I think posting all of the pics is redundant. Taking them is a good idea, and it would be worth noting and posting should there be an anomaly. Barring anomalous results, I don't see the need to clutter the review with multiple TIM mount photos, or even one for that matter.

For instance, you don't see Skinnee or Vapor posting photos of their five different mounts. They may well take them just in case, but we don't see them. I completely agree with thideras that recording them is important, but posting them in the review just adds clutter, IMHO.

[MattNo5ss]

Thid explained it better than I did...lol

Just show any mounts that may have caused abnormalities (good or bad), and a "typical" mount for a base line.

[splat]

i agree that taking pictures of all is the way to go but i was mainly speaking of what to publish and I think publishing more than one picture would be overkill. Thideras is correct that having a record and correlating the pictures to the temps is key to understanding what is happening and why, but I don't think every mount needs to be published and I don't think the results from every mount need to be published. A bad mount is just that, so if you do three runs and one is obviously a bad mount then you do a 4th or 5th to prove it was just a bad mount and not indicative or actual performance. Then you can throw out the high and the low to get the "average performance". If you have 3 mounts, all applied the same, all produce the same spread, and all produce similar temperatures I don't see any need for 3 different pre- and post- pictures.

[thideras]

Why not link them like this.

Here the the pictures for all the different mounts. 1 (http://overclockers.com/1.jpg) - 2 (http://overclockers.com/2.jpg) - 3 (http://overclockers.com/3.jpg) - 4 (http://overclockers.com/4.jpg) - 5 (http://overclockers.com/5.jpg)

BAM, small and very informative. You could also have a small section talking about the spreads for the different mounts. This is an important part of any heatsink test, so why not include it? No reason to show the images, just link to them. This is going to add a lot of depth to the article and I honestly can't think of the last website that included it with every review they did.

Regarding data for different mounts, that takes an extra image. Hell, you could link to it if you wanted to reduce "clutter". This is one of the first images that I reach for as the "average" delta doesn't give me as much detail as I want. Bottom line, include as much data as possible. You could find a way to format it so that it isn't as "cluttered". For example, have the "general" review at the top, a separator in between and the little details below the article. This way you satisfy both audiences. You have the "general" review at the top for people that don't care about details and you have the tons of details that lets you KNOW that the heatsink was tested properly. Saying "Yeah, we tested x, but we don't have proof of it" immediately turns me off on a review and I don't take it seriously. If they leave information out, that is even worse.

EDIT: You could even have two reviews if you want to separate them even more. Have the detailed one not show up on the main page and link to it through the published one on the main article.

[MattNo5ss]

Good idea. IMO, that's the best way to do it. You won't have clutter, but still include all relevant data. As a Mathematician/Research Engineer, I'm all for having the most data as possible. More data never hurts :D

+1 to Thid